Re: [committed] RISC-V: Accept constant operands in la and lla

[Palmer Dabbelt <palmer at sifive dot com>]

On Wed, 20 Jun 2018 21:17:55 PDT (-0700), sebastian.huber@embedded-brains.de wrote:
> On 21/06/18 00:53, Palmer Dabbelt wrote:
> > On Tue, 19 Jun 2018 22:27:25 PDT (-0700), 
> > sebastian.huber@embedded-brains.de wrote:
> > > opcodes/
> > >     PR gas/23305
> > >     * riscv-opc.c (riscv_opcodes): Use new format specifier 'B' for
> > >     la and lla.
> > >
> > > gas/
> > >     PR gas/23305
> > >     * config/tc-riscv.c (riscv_ip): Add format specifier 'B' for
> > >     constants and symbols.
> > >     * testsuite/gas/riscv/lla32.d: New file.
> > >     * testsuite/gas/riscv/lla32.s: Likewise.
> > >     * testsuite/gas/riscv/lla64-fail.d: Likewise.
> > >     * testsuite/gas/riscv/lla64-fail.l: Likewise.
> > >     * testsuite/gas/riscv/lla64-fail.s: Likewise.
> > >     * testsuite/gas/riscv/lla64.d: Likewise.
> > >     * testsuite/gas/riscv/lla64.s: Likewise.
> > > ---
> > >  gas/ChangeLog                        | 13 +++++++++++++
> > >  gas/config/tc-riscv.c                | 11 +++++++++++
> > >  gas/testsuite/gas/riscv/lla32.d      | 19 +++++++++++++++++++
> > >  gas/testsuite/gas/riscv/lla32.s      | 15 +++++++++++++++
> > >  gas/testsuite/gas/riscv/lla64-fail.d |  3 +++
> > >  gas/testsuite/gas/riscv/lla64-fail.l |  2 ++
> > >  gas/testsuite/gas/riscv/lla64-fail.s |  3 +++
> > >  gas/testsuite/gas/riscv/lla64.d      | 20 ++++++++++++++++++++
> > >  gas/testsuite/gas/riscv/lla64.s      | 17 +++++++++++++++++
> > >  opcodes/ChangeLog                    |  6 ++++++
> > >  opcodes/riscv-opc.c                  |  4 ++--
> > >  11 files changed, 111 insertions(+), 2 deletions(-)
> > >  create mode 100644 gas/testsuite/gas/riscv/lla32.d
> > >  create mode 100644 gas/testsuite/gas/riscv/lla32.s
> > >  create mode 100644 gas/testsuite/gas/riscv/lla64-fail.d
> > >  create mode 100644 gas/testsuite/gas/riscv/lla64-fail.l
> > >  create mode 100644 gas/testsuite/gas/riscv/lla64-fail.s
> > >  create mode 100644 gas/testsuite/gas/riscv/lla64.d
> > >  create mode 100644 gas/testsuite/gas/riscv/lla64.s
> >
> > I'm not opposed to this, but did you check with a RISC-V maintainer 
> > before committing it?  In general we try to keep our assembly syntax 
> > compatible between the various assemblers, so any interface addition 
> > should be discussed to make sure it's feasible everywhere. 
>
> Sorry, I should have mentioned that this was pre-approved by Jim Wilson 
> in the referenced PR:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=23305

Ah, OK, sorry I missed that.

> Which other assemblers are there to consider?

At least LLVM, which is spinning up.  There's also a bunch of proprietary tools 
companies, which is what I'm more worried about -- I don't want to have to deal 
with users complaining about undocumented assembly extensions those toolchains 
add so it really wouldn't be fair to have stuff of our own.

> I try to add a documentation of "lla". See also:
>
> https://groups.google.com/a/groups.riscv.org/forum/#!topic/sw-dev/sDQWmHyzHi8

Thanks.  I suppose this should be documented in the GAS manual, though we don't 
have any of the pseudos in there.  Do you mind adding a section to

   gas/doc/c-riscv.texi

for pseduo ops that contains lla?  We can add the rest later...

Additionally, we've been collecting all the standard interfaces in RISC-V 
specific manuals.  Do you mind adding a bit here about lla (and then modifying 
the la bit to include that it can take constants)

   https://github.com/riscv/riscv-asm-manual/blob/master/riscv-asm.md

I also left a comment in the middle of the source about "lla 0xFFFFFFFF" on 
RV64I, which I think should be supported.

Thanks for the patch, and sorry for making so much more work :)